Project Title:

A Recurrent Neural Network/Expert System Approach to Automated Task Planning and Resource Allocation

94-1	06.02	3474		

A Recurrent Neural Network/Expert System Approach to Automated Task 
Planning and Resource Allocation

Abstract:

In the proposed Phase I study we will develop and demonstrate the 
validity of a recurrent neural-network-based general task 
scheduling and resource allocation software. A hybrid methodology 
that combines expert system techniques with a recurrent neural 
network combinatorial optimization module is proposed. The 
recurrent neural network is used for solving the task scheduling 
and resource allocation problem. Expert systems automate the 
generation of the neural network architecture from user 
specifications, analyze the neural network states after convergence 
and provide expert advice to the user on how to improve the 
resulting solution. The proposed study consists of five tasks: 1) 
the design of the recurrent-neural-network-based constrained 
optimizer; 2) the specification and implementation of an automated 
neural network designer; 3) the development of a network state 
interpreter and a diagnostic module; 4) validation of a working 
limited-scope prototype; and 5) specification of design 
requirements for a full-scale prototype.

There exists a number of commercial applications for an efficient 
task scheduling and resource optimization system. These include 
flexible manufacturing systems, general project management systems, 
real-time distributed processing systems and large network 
management

Key Words

Charles River Analytics	   
55 Wheeler Street
Cambridge, MA 02138